Abstract:The mixed quantum-classical dynamical simulation is essential for studying nonadiabatic phenomena in photophysics and photochemistry. In recent years, many machine learning models have been developed to accelerate the time evolution of the nuclear subsystem. Herein, we implement long short-term memory (LSTM) networks as a propagator to accelerate the time evolution of the electronic subsystem during the fewest-switches surface hopping (FSSH) simulations. A small number of reference trajectories...

chemistry, physical,physics, atomic, molecular & chemical,nanoscience & nanotechnology,materials science, multidisciplinary

Fewest-Switches Surface Hopping with Long Short-Term Memory Networks